Athens

Cursor for Writing: Why Inline Suggestions Beat Chat Rewrites

- Moritz Wallawitsch

If you've used Cursor for coding, you know the feeling: you highlight some code, ask the AI to refactor it, and see a precise diff appear inline. Green for additions, red for deletions. Accept, reject, or tweak each change individually. Your flow never breaks because the AI meets you where you're already working.

Now compare that to how most people use AI for writing: open ChatGPT in a separate tab, paste a paragraph, ask it to rewrite, squint at the result trying to figure out what changed, paste it back, fix the formatting. Repeat fifty times.

Writing deserves the Cursor treatment. Here's what that looks like.

The Problem with Chat-Based Editing

Chat interfaces are great for generating text from scratch. They're terrible for editing existing text. The reason is simple: editing requires context and precision. You need the AI to understand your full document, target a specific section, and show you exactly what it changed.

Chat gives you none of that:

  • No document context. You paste a paragraph, but the AI doesn't see the rest of your document. It can't maintain consistency in tone, terminology, or argument structure.
  • No visibility into changes. The AI returns a complete rewrite. What did it actually change? You have to read both versions line by line to find out.
  • All-or-nothing. If you like 80% of the rewrite but not the other 20%, you're stuck manually merging. There's no way to accept some changes and reject others.
  • Context window decay. After 30-40 messages, the AI starts forgetting earlier instructions. Long editing sessions degrade in quality.

These aren't bugs in ChatGPT. They're fundamental limitations of using a chat interface for document editing.

What Inline Suggestions Look Like

The alternative is what developers have had for years: the AI edits your document directly, and you review each change before it takes effect.

In practice, this means:

  1. You select text (or just describe what you want changed in a chat sidebar)
  2. The AI reads your entire document for context
  3. It produces targeted edits - not a full rewrite, but specific text replacements
  4. Each edit appears as an inline suggestion: red strikethrough for deletions, green highlight for insertions
  5. You accept or reject each edit individually with one click

This is fundamentally different from chat. The AI proposes; you decide. You stay in your document the entire time. And you can see exactly what the AI is doing at a granular level.

Why This is Better for Writers

Writers aren't looking for a ghostwriter. They're looking for a sharp editor who makes targeted suggestions. The best human editors don't rewrite your work - they mark up specific passages with comments like "tighten this," "wrong word here," or "this paragraph buries the lead."

Inline AI suggestions replicate this editorial relationship. The AI is your first reader, catching awkward phrasing, redundant sentences, and unclear arguments. But you retain full control. Every change goes through you.

This matters for voice. The biggest complaint about AI writing tools is that they make everything sound the same - that generic, slightly enthusiastic, overly polished AI tone. When you accept and reject individual changes, you preserve your voice while benefiting from AI's ability to spot issues you might miss.

The Technical Insight

Cursor works so well because code is plain text. The AI can read it directly, produce text diffs, and the editor can display those diffs inline.

The same approach works for writing - if your editor stores documents in a text-based format. This is why markdown is the ideal storage format for AI-native editors. The AI reads markdown, produces targeted text replacements, and the editor renders the diff in the WYSIWYG view. No format conversion, no block mutation APIs, no lossy translation.

What's Missing from Most AI Writing Tools

Most AI writing tools in 2026 still treat AI as a separate mode. You're either writing or you're prompting. Notion has an AI sidebar. Google Docs has "Help me write." These feel bolted on because they are - the editor wasn't designed around AI from the start.

The Cursor model shows what's possible when AI is a first-class citizen in the editor. You don't switch modes. You don't leave your document. You just write, and when you want AI help, it's right there - making suggestions you can accept or reject without breaking your flow.

That's the experience we're building with Athens. A markdown WYSIWYG editor where AI edits appear as inline suggestions - Cursor-style diffs for prose. Because writers deserve the same tools that developers have had for years.